diff options
| -rw-r--r-- | Codes/checkpoints/download_pretrains.sh | 11 | ||||
| -rw-r--r-- | Codes/constant.py | 6 | ||||
| -rw-r--r-- | Codes/models/download_pretrains.sh | 11 | ||||
| -rw-r--r-- | Codes/requirements.txt | 2 | ||||
| -rw-r--r-- | Codes/utils.py | 86 | ||||
| -rw-r--r-- | README.md | 14 |
6 files changed, 32 insertions, 98 deletions
diff --git a/Codes/checkpoints/download_pretrains.sh b/Codes/checkpoints/download_pretrains.sh new file mode 100644 index 0000000..a85f19c --- /dev/null +++ b/Codes/checkpoints/download_pretrains.sh @@ -0,0 +1,11 @@ +#!/bin/bash + +echo "Downloading trained models on ped1, ped2 and avenue datasets ....." + +wget "https://ofhz9a.bn.files.1drv.com/y4mdc9K-lh3tfqBXiG-rSR04JARM20at-t2AtTo-7fUC-fadMB_x255o35v0J-YV4bnMW9m9XscVOKkITHI2bZLrgxZJJKXI4QEVDsi61KvsLQxI42elYWmm01F2kjI94onmbRrFYai7CkVNUspBgscY2vvEfd2c3qE2A_bcTW-Cp_6hBKpPEQClmwlT2QqTy-UwuzCmjyFfOrHqKeGkqtadQ/pretrains.tar.gz" +tar -xvf pretrains.tar.gz +rm pretrains.tar.gz + +echo "Download pretrains successfully..." + + diff --git a/Codes/constant.py b/Codes/constant.py index f6f78df..b4fc7e3 100644 --- a/Codes/constant.py +++ b/Codes/constant.py @@ -92,7 +92,7 @@ const.EVALUATE = args.evaluate # network constants const.HEIGHT = 256 const.WIDTH = 256 -const.FLOWNET_CHECKPOINT = 'models/pretrains/flownet-SD.ckpt-0' +const.FLOWNET_CHECKPOINT = 'checkpoints/pretrains/flownet-SD.ckpt-0' const.FLOW_HEIGHT = 384 const.FLOW_WIDTH = 512 @@ -136,9 +136,9 @@ if args.snapshot_dir: const.SNAPSHOT_DIR = args.snapshot_dir print(const.SNAPSHOT_DIR) else: - const.SNAPSHOT_DIR = get_dir(os.path.join('models', const.SAVE_DIR + '_' + args.snapshot_dir)) + const.SNAPSHOT_DIR = get_dir(os.path.join('checkpoints', const.SAVE_DIR + '_' + args.snapshot_dir)) else: - const.SNAPSHOT_DIR = get_dir(os.path.join('models', const.SAVE_DIR)) + const.SNAPSHOT_DIR = get_dir(os.path.join('checkpoints', const.SAVE_DIR)) if args.summary_dir: const.SUMMARY_DIR = get_dir(os.path.join('summary', const.SAVE_DIR + '_' + args.summary_dir)) diff --git a/Codes/models/download_pretrains.sh b/Codes/models/download_pretrains.sh deleted file mode 100644 index d2a7da8..0000000 --- a/Codes/models/download_pretrains.sh +++ /dev/null @@ -1,11 +0,0 @@ -#!/bin/bash - -echo "Downloading trained models on ped1, ped2 and avenue datasets ....." - -wget "https://ofhz9a.bn.files.1drv.com/y4mljH6-fb1fLS6B9cJg5kst-wanKhmgoA8mWk_8y8atGlL4SQqEqZRljA8LuLkxkvm-36PArbAsx-uYKxzfUX1s8otO8h0Rv7tag9Agx1h8_RsTPhQo-aft0QSybwpvcvOkegqfLVg2UJjTzIsxgizIYCVDIpunmpasyB3t4mchRGZetZQGiKP4fS2Vfb4c3tj1jxzpN61Pmou9QaB3VAG0w/pretrains.tar.gz" -tar -xvf pretrains.tar.gz -rm pretrains.tar.gz - -echo "Download pretrains successfully..." - - diff --git a/Codes/requirements.txt b/Codes/requirements.txt index 91d2206..8ab8d6c 100644 --- a/Codes/requirements.txt +++ b/Codes/requirements.txt @@ -2,7 +2,7 @@ numpy==1.14.1 scipy==1.0.0 matplotlib==2.1.2 tensorflow==1.4.1 -tensorflow_gpu==1.4.1 +tensorflow-gpu==1.4.1 Pillow==5.0.0 pypng==0.0.18 scikit_learn==0.19.1 diff --git a/Codes/utils.py b/Codes/utils.py index efeab8e..8466d08 100644 --- a/Codes/utils.py +++ b/Codes/utils.py @@ -10,6 +10,15 @@ rng = np.random.RandomState(2017) def np_load_frame(filename, resize_height, resize_width): + """ + Load image path and convert it to numpy.ndarray. Notes that the color channels are BGR and the color space + is normalized from [0, 255] to [-1, 1]. + + :param filename: the full path of image + :param resize_height: resized height + :param resize_width: resized width + :return: numpy.ndarray + """ image_decoded = cv2.imread(filename) image_resized = cv2.resize(image_decoded, (resize_width, resize_height)) image_resized = image_resized.astype(dtype=np.float32) @@ -20,7 +29,7 @@ def np_load_frame(filename, resize_height, resize_width): class DataLoader(object): def __init__(self, video_folder, resize_height=256, resize_width=256): self.dir = video_folder - self.videos = {} + self.videos = OrderedDict() self._resize_height = resize_height self._resize_width = resize_width self.setup() @@ -83,15 +92,6 @@ class DataLoader(object): return np.concatenate(batch, axis=2) - # def get_video_clips(self, video_name, start, end): - # video_idx = np.arange(start, end) - # video_clip = np.empty(shape=[self._resize_height, self._resize_height, 3*len(video_idx)], dtype=np.float32) - # for idx, v_idx in enumerate(video_idx): - # filename = self.videos[video_name]['frame'][v_idx] - # video_clip[..., idx*3:(idx+1)*3] = np_load_frame(filename, self._resize_height, self._resize_width) - # - # return video_clip - def log10(t): """ @@ -130,46 +130,6 @@ def psnr_error(gen_frames, gt_frames): return tf.reduce_mean(batch_errors) -def sharp_diff_error(gen_frames, gt_frames, channels=3): - """ - Computes the Sharpness Difference error between the generated images and the ground truth - images. - - @param gen_frames: A tensor of shape [batch_size, height, width, 3]. The frames generated by the - generator model. - @param gt_frames: A tensor of shape [batch_size, height, width, 3]. The ground-truth frames for - each frame in gen_frames. - @param channels: The number of channels, 3 is RGB and 1 is Gray, default is 3. - - @return: A scalar tensor. The Sharpness Difference error over each frame in the batch. - """ - shape = tf.shape(gen_frames) - num_pixels = tf.to_float(shape[1] * shape[2] * shape[3]) - - # gradient difference - # create filters [-1, 1] and [[1],[-1]] for diffing to the left and down respectively. - # TODO: Could this be simplified with one filter [[-1, 2], [0, -1]]? - pos = tf.constant(np.identity(channels), dtype=tf.float32) - neg = -1 * pos - filter_x = tf.expand_dims(tf.stack([neg, pos]), 0) # [-1, 1] - filter_y = tf.stack([tf.expand_dims(pos, 0), tf.expand_dims(neg, 0)]) # [[1],[-1]] - strides = [1, 1, 1, 1] # stride of (1, 1) - padding = 'SAME' - - gen_dx = tf.abs(tf.nn.conv2d(gen_frames, filter_x, strides, padding=padding)) - gen_dy = tf.abs(tf.nn.conv2d(gen_frames, filter_y, strides, padding=padding)) - gt_dx = tf.abs(tf.nn.conv2d(gt_frames, filter_x, strides, padding=padding)) - gt_dy = tf.abs(tf.nn.conv2d(gt_frames, filter_y, strides, padding=padding)) - - gen_grad_sum = gen_dx + gen_dy - gt_grad_sum = gt_dx + gt_dy - - grad_diff = tf.abs(gt_grad_sum - gen_grad_sum) - - batch_errors = 10 * log10(1 / ((1 / num_pixels) * tf.reduce_sum(grad_diff, [1, 2, 3]))) - return tf.reduce_mean(batch_errors) - - def diff_mask(gen_frames, gt_frames, min_value=-1, max_value=1): # normalize to [0, 1] delta = max_value - min_value @@ -197,31 +157,5 @@ def save(saver, sess, logdir, step): print('The checkpoint has been created.') -# if __name__ == '__main__': -# os.environ['CUDA_DEVICES_ORDER'] = "PCI_BUS_ID" -# os.environ['CUDA_VISIBLE_DEVICES'] = '0' -# -# data_loader = DataLoader('/home/liuwen/ssd/datasets/avenue/training/frames') -# dataset, epoch_size = data_loader(10, 4, 1, 3, 1) -# -# # debug -# iteration = dataset.make_one_shot_iterator() -# batch_video_clip_tensor = iteration.get_next() -# -# config = tf.ConfigProto() -# config.gpu_options.allow_growth = True -# with tf.Session(config=config) as sess: -# # batch_video_clip = sess.run(next(it)) -# -# for i in range(100): -# batch_video_clip = sess.run(batch_video_clip_tensor) -# # print(batch_video_clip.shape) -# -# for vid, video_clip in enumerate(batch_video_clip): -# for fid, frame in enumerate(video_clip): -# print(i, vid, fid) -# cv2.imshow('visualization', frame + 0.5) -# cv2.waitKey(100) - @@ -1,5 +1,5 @@ # Future Frame Prediction for Anomaly Detection -- A New Baseline -This repo is the official open source of [Future Frame Prediction for Anomaly Detection -- A New Baseline, CVPR 2018](https://arxiv.org/pdf/1712.09867.pdf) by Wen Liu, Weixinluo, Dongze Lian and Shenghua Gao. +This repo is the official open source of [Future Frame Prediction for Anomaly Detection -- A New Baseline, CVPR 2018](https://arxiv.org/pdf/1712.09867.pdf) by Wen Liu, Weixin Lluo, Dongze Lian and Shenghua Gao. A **demo** is shown in *https://www.youtube.com/watch?v=M--wv-Y_h0A*. It is implemented in tensorflow. Please follow the instructions to run the code. ## 1. Installation (Anaconda with python3.6 installation is recommended) @@ -8,7 +8,7 @@ A **demo** is shown in *https://www.youtube.com/watch?v=M--wv-Y_h0A*. It is impl numpy==1.14.1 scipy==1.0.0 matplotlib==2.1.2 -tensorflow_gpu==1.4.1 +tensorflow-gpu==1.4.1 tensorflow==1.4.1 Pillow==5.0.0 pypng==0.0.18 @@ -37,7 +37,7 @@ cd Data ## 3. Testing on saved models * Download the trained models ```shell -cd models +cd checkpoints ./download_pretrains.sh ``` * Running the sript (as ped2 and avenue datasets for examples) and cd into **Codes** folder at first. @@ -45,14 +45,14 @@ cd models python inference.py --dataset ped2 \ --test_folder ../Data/ped2/testing/frames \ --gpu 1 \ - --snapshot_dir models/pretrains/ped2 + --snapshot_dir checkpoints/pretrains/ped2 ``` ```shell python inference.py --dataset avenue \ --test_folder ../Data/avenue/testing/frames \ --gpu 1 \ - --snapshot_dir models/pretrains/avenue + --snapshot_dir checkpoints/pretrains/avenue ``` @@ -111,8 +111,8 @@ Open the browser and type **https://ip:10086**. Following is the screen shot of   -Since all frames are loaded into BGR channels in training and testing, the visualized images look different from RGB channels. - +Since the models are trained in BGR image color channels, the visualized images in tensorboard look different from RGB channels. +In the demo, we change the output images from BGR to RGB. ## Notes The flow loss (temporal loss) module is based on [a TensorFlow implementation of FlowNet2](https://github.com/sampepose/flownet2-tf). Thanks for their nice work. |